Research Of Humming Retrieval Based On Music Classification

Humming retrieval is a method based on music content.Using this method, user only need to remember some fragment of the music instead of the name of music, music singer, composer, songwriter and a series of complex information.This paper proposed a new humming retrievalmethod based on music classification. The method mainly studied the following contents:①Training each set of music files and establish the hidden Markov model of each category.②Extract the physic features of humming signal as the input of the trained hidden Markov model and classify the humming signal.③Extract the classified humming signal melody feature then use it for similarity matching with the sequence data in the music feature data set. Then return the result of similarity matching.According to the need of the paper, four jobs would be done.1) Study on the preprocessing of the humming signal.2) Study on the feature extracted.3) Study on the melody match.4) Study on the music classification.This papermainly do two comparative experiments. One is the comparison between retrieval experiments based on HMM and experiments based on HMM joined with music classification. The other one is the comparison between the experiments based on dynamic time warping(DTW) and experiments based on DTW joined with music classification. The comparing and analyzing the results of the experiments showed that humming retrieval based on music classification can some extent shorten the average search time.